Abstract:
PURPOSE: A processing device, a PE control device, and a method thereof are provided to support predication execution in a SIMD(Single Command Multiple Data) machine by the state of PE is changed according to the conditional information of sleep command. CONSTITUTION: A PE(Processing Element)(200) performs calculation according to control signals. A PE controller(210) transmits control signals outputted based on external command to the PE. When the command is a sleep command including condition information and a sleep cycle value, the PE controller compares the condition information of the sleep command with the former calculation result performed by the PE. When the calculation result is matched in the condition of the condition information of the sleep command, the PE controller blocks the electrical transmission of the control signal according to the command for the cycle of the sleep cycle value.
Abstract:
본 발명은 종래의 재구성 어레이 구조와 거의 비슷한 하드웨어 크기를 가지면서도 부동 소수점 연산을 비교적 빠르게 수행할 수 있고, 저전력 및 가격대비 고성능비 특성을 갖는 재구성 어레이 프로세서(Reconfigurable Array Processor) 및 그 프로세서를 포함한 멀티미디어 플랫폼을 제공한다. 그 재구성 어레이 프로세서는 2개 이상의 정수 연산 프로세싱 요소(processing element: PE)가 결합하여 이루어진 부동 소수점 유닛(Floating-Point Unit: FPU)-PE가 다수 개 배열된 PE 어레이(array); 상기 PE 어레이의 행 또는 열 별로 PE 연산 및 PE 간의 데이터 통신을 제어하는 컨텍스트(context)가 저장되는 구성 캐쉬(Configurable Cache); 및 상기 PE 어레이가 수행한 중간 결과를 저장하는 캐시 메모리인 프레임 버퍼(Frame buffer);를 포함하고, 상기 FPU-PE를 이용하여 부동-소수점(floating point) 연산을 수행할 수 있다.
Abstract:
A floating point unit-processing element structure, a reconfigurable array processor comprising the floating point unit-processing element structure, and a multimedia platform comprising the reconfigurable array processor are provided to rapidly a floating point operation while having almost similar hardware size to an existing reconfigurable array processor by using an existing integer operation processing element. In a processing element array(100), a plurality of PE(Processing Element)s(120) are arranged. The PE comprises an ALU(Arithmetic and Logic Unit) for performing an integer operation by receiving two input values. An FPU-PE(Floating Point Unit-Processing Element) is formed by combining at least two PEs to support the floating operation. In the processing element array, the PEs are connected in a mesh structure to exchange data between the PEs. Each PE array performs each integer operation during integer operation. The FPU-PE is formed by combining at least two PEs during the floating point operation to perform the floating point operation.
Abstract:
A semiconductor memory device including a non-volatile memory and a cache memory and a computer system comprising the same are disclosed. The cache memory according to an embodiment of the present invention comprises: a first storage area including a plurality of data blocks for storing data; and a second storage area for including m number of ECC blocks corresponding to n number of data blocks, wherein each of the first and the second storage areas includes non-volatile memory cells. At least one ECC block stores first information indicating an ECC parity for correcting an error bit of the corresponding data block and a position of the corresponding data block.
Abstract:
PURPOSE: A network-on-chip system including an active memory processor is provided to improve performance of a parallel application and has a surface overhead which is smooth in a network interface of a memory tile. CONSTITUTION: An PE(Processing Element)(110) requests an active memory operation with the sharing memory for reducing an access latency of a shared memory(130). An active memory processor(122) is connected through the PEs and a network and stores a code for processing a custom transaction according to the request of the active memory operation. The active memory processor performs calculation about an address or a data saved in shared cache memory or a shared memory based on code and transmits an operation result with PEs.
Abstract:
본 발명은 네트워크-온-칩에 관한 것으로, 메모리 주변에 위치하고 능동 메모리 연산을 실행하는 능동 메모리 프로세서라고 불리는 프로세서를 구현함으로써, 온-칩 네트워크에서 다수의 메모리 액세스 트랜잭션 및 관련 로컬 프로세싱 엘리먼트 계산을 더 적은 수의 하이-레벨 트랜잭션 및 메모리-근접 계산으로 대체할 수 있는 효과가 있다.
Abstract:
PURPOSE: An in-out data address generation circuit and a method of continuous flow memory-based FFT processor using multiple butterflies are provided to selectively control the number of butterflies included in the consecutive processing FFT processor without increasing memory usage. CONSTITUTION: An operation unit(101) comprises a plurality of butterflies processing the butterfly operation. An input/output interface(104) processes data input/output from the multiple bank memory. An input data exchange circuit(105) supplies the butterfly input from the multiple bank memories to the operation unit. An output data exchange circuit(106) supplies the butterfly output from the operation unit to multiple bank memories.
Abstract:
PURPOSE: A variable-length decoding apparatus and a method for rapidly inputting the input stream encoded to the variable length encoding type are provided to perform quick decoding process by detecting symbols belonging to sub-group based on a lookup table search. CONSTITUTION: A first operating unit(16) decides symbols included in the upper group or sub-group. If the symbols are included in sub-group, the first operation unit detects the lookup table information corresponding to the subgroup in which symbols are included. If the lookup table information is received, a second operating unit(17) refers to the look up table corresponding to the lookup table information.